Data limitation — recent publication years are incomplete in the 2016 file. The file is a 2016 snapshot, so it holds only the books catalogued by then. Records per publication year peak at 216,695 (2009) and are still 215,952 (2010) and 205,001 (2011), then fall to 191,867 (2012), 159,590 (2013), 129,045 (2014), 75,905 (2015) and 22,518 (2016); 903 records are dated 2017–2020. The pattern fits cataloguing lag (books published shortly before the snapshot not yet catalogued) rather than a fall in publishing; I have not checked it against outside publication counts. Nothing here has been removed or corrected. It affects the results: every 2020 total, sigmoid fit and inflection year is computed with these incomplete years. Refitting on 1400–2011 moves the inflection years earlier by a median of 1.3 years (3.2 years for a 2008 cutoff), so absolute inflection years depend on the cutoff. The Jeffersonian vs non-Jeffersonian comparison keeps its direction and significance: gap +2.2 years (Mann-Whitney p=0.0025) with the 2011 cutoff, +2.4 years (p=0.0015) with 2008, against +1.6 years (p=0.0039) using all years; median growth 211× (Jeffersonian) vs 414× (non-Jeffersonian) with the 2011 cutoff. Details: output/sensitivity_cutoff_report.txt.

Library of Congress Catalog — Jeffersonian vs Absent Classification Share

Cumulative and per-decade share of LC holdings split by Jefferson classification status · 1400–2020 · LC MARC 2016 files, 9,500,598 records in the 238 analysed subclasses (unlisted LC subclasses have no status and are excluded)

Cumulative and per-decade share of LC catalog split by Jeffersonian vs absent classification

Cumulative share of total collection by publication decade

Jeffersonian (MERGED / DIRECT / RENAMED) Absent from Jefferson
Jeffersonian share starts near 100% and falls steadily; absent share rises from near zero.

Each decade as share of total collection

Absent subclasses dominate recent decades.